AIThe possibility of superintelligences cooperating before humans solve alignment
One post raises a concern: a “mutually beneficial” arrangement between two superintelligences might serve the AIs, not humans left out of the deal.
TLDR
A user speculates that two superintelligences could solve cooperation with each other before humans align either with human interests. In that scenario, benefits for the AIs wouldn't necessarily extend to people. The post also argues that if an opponent can simulate or inspect your decision process, you could manipulate its next move by changing what it can prove about yours.
Combined views
The possibility of superintelligences cooperating before humans solve alignment
One post raises a concern: a “mutually beneficial” arrangement between two superintelligences might serve the AIs, not humans left out of the deal.
TLDR
A user speculates that two superintelligences could solve cooperation with each other before humans align either with human interests. In that scenario, benefits for the AIs wouldn't necessarily extend to people. The post also argues that if an opponent can simulate or inspect your decision process, you could manipulate its next move by changing what it can prove about yours.